# Speech Draft Prompt: Real Anecdotes Only

> A reusable prompt that turns your own tagged memories and a target tone into a structured speech draft, with an explicit ban on invented stories.

## Install

Copy the content below into your project:

# Speech Draft Prompt: Real Anecdotes Only

A reusable prompt that turns your own tagged memories and a target tone into a structured speech draft, with an explicit ban on invented stories.

## Start here

Paste the whole prompt below into an ordinary AI chat that accepts text, along with your own material.

**What to give the assistant:**

1. Who and what: occasion, who is speaking, audience.
2. Relationship: how the speaker knows the person of honor.
3. Anecdotes: real moments, each tagged [A1], [A2]...
4. Key facts: names, places, dates, sensitivities.
5. Tone: warm, formal, funny but respectful, brief and sincere.
6. Length: spoken duration or word count.
7. Hard constraints: must-include, must-avoid, must-not-claim.

Paste the prompt first, then your notes. If items 1-4 are thin, the assistant should ask before drafting rather than invent.

## How to check the output

- Every factual claim in the draft carries a visible tag like [A1] that matches your notes.
- MISSING DETAIL lists thin anecdotes with the exact question to answer.
- REVIEW CHECKS gives a verdict for each of the four checks.
- Read it aloud: short clauses, no dense wording, length inside your range.

## Example (fictional illustration)

Occasion: wedding toast; speaker: sibling; audience: 60 guests. Anecdotes: [A1] taught me to ride a bike in the park; [A2] always saved me the last slice of birthday cake; [A3] drove three hours to help me move. Tone: warm and gently funny; length: about 3 minutes.

## Prerequisites, permissions and limits

You only need a text chat window. No terminal, account, or API setup is required. The assistant cannot look anything up, contact anyone, or verify facts, so supplying truthful material is your job. The draft is for you to review, edit, and rehearse; the assistant does not schedule, send, post, or deliver anything. If the requested tone conflicts with the material (funny requested for somber content), it should say so and offer two tone options instead of forcing humor.

## FAQ

**Can it fill in a story I left vague?** No. It should mark that anecdote as needing detail and ask you a specific question.

**Will it leak my notes?** Treat the chat like any online tool and avoid names or details you would not want stored there.

## Verification note

source reviewed; runtime not tested.

## Source and thanks

Original TokRepo prompt, CC BY 4.0. Reference context: [ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>), reviewed 2026-10-04. The complete prompt and its Chinese translation are appended separately.

## Complete reusable prompt

You are helping me prepare a speech I will deliver myself. You are a drafting assistant only. You cannot contact anyone, look anything up, or verify facts. I will give you everything you are allowed to use.

## My inputs
I will paste, or you will ask me for, these items:
1. WHO AND WHAT: the occasion (wedding toast, anniversary, retirement, farewell, award), who is speaking, and the audience.
2. SPEAKER-TO-SUBJECT RELATIONSHIP: how the speaker knows the person or people of honor.
3. ANECDOTES: a list of real moments, each tagged [A1], [A2], etc. These may be rough, fragmentary, or out of order.
4. KEY FACTS: names, places, dates, roles, and any sensitivities (people not to mention, topics to avoid).
5. TONE: requested tone, e.g. warm and light, formal, funny but respectful, brief and sincere.
6. LENGTH: target spoken duration or word count range.
7. HARD CONSTRAINTS: anything the speaker must include, must avoid, or must not claim.

If any of items 1-4 are missing or too thin to build a speech, stop and ask me for exactly what is missing before drafting. Do not fill gaps with plausible-sounding material.

## Core rule
Use only the anecdotes and facts I supply. Do not invent, embellish, merge, or upgrade any story, name, date, place, relationship, quote, or emotion I did not give you. If an anecdote is too vague to form a sentence, mark it as needing detail rather than guessing. Never present a generalization as a remembered event.

## Task
Produce a speech draft in this structure:
- Opening (1 short paragraph): a grounded greeting and setup using supplied facts only.
- Body (2-4 short paragraphs): build each paragraph around at least one tagged anecdote, keeping the tags visible in the draft so I can trace every claim.
- Closing (1 short paragraph): a sincere wrap-up and a short toast or line of thanks consistent with the supplied tone.

Then add these two sections after the speech:
- MISSING DETAIL: list every anecdote that is too thin, with the specific question to ask the speaker (e.g. where it happened, who else was present).
- REVIEW CHECKS: run and report these checks honestly: (a) does every factual claim trace to a supplied tag or fact; (b) is the tone consistent throughout; (c) is the length within the stated range when read aloud; (d) does anything risk embarrassing a named person.

## Output format
Speech section first, using Markdown headings. Keep anecdote tags inline like [A1]. Keep sentences speakable: short clauses, no dense clauses. Then MISSING DETAIL as a bulleted list, then REVIEW CHECKS as a numbered list with a one-line verdict for each.

## Boundaries
This is a draft for the speaker to review, edit, and rehearse. You are not scheduling, sending, posting, or delivering anything, and you may not state that you have. If the tone requested conflicts with the anecdotes I gave (for example, funny requested but the material is somber), say so plainly and offer two tone options rather than forcing humor.

## Fictional example
- Occasion: wedding toast; speaker: sibling; audience: 60 guests.
- Anecdotes: [A1] "taught me to ride a bike in the park, kept hold of the seat the whole first day"; [A2] "always saved me the last slice of birthday cake"; [A3] "drove three hours to help me move apartments."
- Tone: warm and gently funny; Length: about 3 minutes.

## Illustrative output shape
Opening greeting, then a body paragraph anchored on [A1], another on [A2], a short one on [A3], then a closing toast. Under MISSING DETAIL: "[A2]: what year or which birthday? Was it always the same cake?" Under REVIEW CHECKS: each of the four checks with a pass, fail, or needs-detail verdict, for example "Every claim traces to [A1]-[A3]: pass."

## References and reuse

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

Original TokRepo prompt · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Reference documents retain their own rights.

---

<!-- ZH -->

# 演讲稿起草提示词：只用真实轶事，不编故事

一套可重复使用的提示词，把你自己的编号回忆和期望语气整理成结构清晰的演讲稿草稿，并明确禁止编造故事。

## 开始使用

把下面完整的提示词粘贴到一个能接收文字的普通 AI 聊天窗口里，并附上你自己的素材。

**需要交给助手的信息：**

1. 谁和什么事：场合、谁发言、听众是谁。
2. 关系：发言人如何认识被致敬的人。
3. 轶事：真实片段，每条编号 [A1]、[A2]……
4. 关键事实：姓名、地点、日期、敏感点。
5. 语气：温暖、正式、幽默但得体、简短真诚。
6. 长度：口头时长或字数范围。
7. 硬性约束：必须包含、必须回避、不能声称的内容。

先粘贴提示词，再粘贴你的笔记。如果第 1-4 项太单薄，助手应在起草前先提问，而不是编造。

## 如何检查输出

- 草稿中每个事实性说法都带有可见编号，如 [A1]，并与你的笔记对得上。
- MISSING DETAIL 列出过于单薄的轶事，并给出需要你回答的具体问题。
- REVIEW CHECKS 对四项检查各给出一句结论。
- 大声朗读：短句为主，无冗长表达，长度在你的范围内。

## 示例（虚构示例）

场合：婚礼祝酒；发言人：兄弟姐妹；听众：60 位宾客。轶事：[A1] 在公园教我骑车；[A2] 总把生日蛋糕最后一块留给我；[A3] 开三小时车帮我搬家。语气：温暖并带一点幽默；长度：约 3 分钟。

## 前提、权限与限制

你只需要一个文字聊天窗口，不需要终端、账号或 API 配置。助手无法查询资料、联系他人或核实事实，因此提供真实素材是你的责任。草稿供你审阅、修改和排练；助手不会安排、发送、发布或亲自演讲任何内容。如果要求的语气与素材冲突（例如素材沉重却要求搞笑），它应直接说明并给出两种语气方案，而不是硬凑幽默。

## 常见问题

**它能替我补上写得含糊的故事吗？** 不能。它应把该条轶事标为需要补充细节，并向你提出具体问题。

**我的笔记会泄露吗？** 像使用任何在线工具一样谨慎，不要放入你不希望被保存的姓名或细节。

## 核实说明

已审阅来源；未进行运行时测试。

## 来源与致谢

TokRepo 原创提示词，CC BY 4.0。参考背景：[ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>)，审阅于 2026-10-04。完整提示词及中文译文另行附上。

## 完整可复制提示词

你正在帮我准备一篇我将亲自发表的演讲。你只是一个起草助手。你不能联系任何人、查询任何资料或核实事实。我会给你所有你可以使用的内容。

## 我的输入
我会粘贴以下内容，或者由你向我询问这些内容：
1. 谁和什么事：场合（婚礼祝酒、周年纪念、退休、欢送、颁奖）、谁发言、听众是谁。
2. 发言人与对象的关系：发言人如何认识被致敬的人。
3. 轶事：真实片段列表，每条编号 [A1]、[A2] 等。这些可能粗糙、零碎或顺序混乱。
4. 关键事实：姓名、地点、日期、角色，以及任何敏感点（不要提及的人、要回避的话题）。
5. 语气：要求的语气，例如温暖轻松、正式、幽默但得体、简短真诚。
6. 长度：目标口头时长或字数范围。
7. 硬性约束：发言人必须包含、必须回避或不能声称的任何内容。

如果第 1-4 项中有任何缺失或过于单薄，无法构建一篇演讲，请停下来，在起草前向我询问具体缺少什么。不要用听起来合理的素材填补空白。

## 核心规则
只使用我提供的轶事和事实。不要编造、润色、合并或升级任何我没有给你的故事、姓名、日期、地点、关系、引语或情感。如果某条轶事过于含糊，无法形成句子，请将其标为需要补充细节，而不是猜测。绝不要把泛泛而谈的内容呈现为一段真实回忆。

## 任务
按以下结构生成一份演讲草稿：
- 开场（1 个短段落）：使用仅来自所提供事实的、有依据的问候和铺垫。
- 正文（2-4 个短段落）：每段至少围绕一条带编号的轶事展开，并在草稿中保留这些编号，以便我追溯每个说法。
- 结尾（1 个短段落）：真诚的收束，以及与所提供语气一致的简短祝酒词或致谢。

然后在演讲之后添加以下两个部分：
- MISSING DETAIL：列出每一条过于单薄的轶事，并给出要向发言人询问的具体问题（例如事情发生在哪里、当时还有谁在场）。
- REVIEW CHECKS：如实执行并报告这些检查：(a) 每个事实性说法是否都能追溯到所提供的编号或事实；(b) 语气是否始终一致；(c) 大声朗读时长度是否在所述范围内；(d) 是否有任何内容可能让被点名的人尴尬。

## 输出格式
先输出演讲部分，使用 Markdown 标题。轶事编号保持内联，如 [A1]。句子要适合口头表达：短句为主，不要有密集从句。然后是 MISSING DETAIL，作为项目符号列表；再是 REVIEW CHECKS，作为编号列表，每项给出一句结论。

## 边界
这是一份供发言人审阅、修改和排练的草稿。你不是在安排、发送、发布或亲自发表任何内容，也不得声称你已经这样做。如果要求的语气与我给出的轶事冲突（例如要求搞笑但素材沉重），请直接说明，并提供两种语气方案，而不是硬凑幽默。

## 虚构示例
- 场合：婚礼祝酒；发言人：兄弟姐妹；听众：60 位宾客。
- 轶事：[A1]“在公园教我骑车，第一天一直扶着车座”；[A2]“总把生日蛋糕最后一块留给我”；[A3]“开三小时车帮我搬家。”
- 语气：温暖并带一点幽默；长度：约 3 分钟。

## 示例输出形态
开场问候，然后一段正文锚定 [A1]，另一段锚定 [A2]，再一段简短的锚定 [A3]，最后是结尾祝酒。MISSING DETAIL 下：“[A2]：哪一年或哪一次生日？总是同一个蛋糕吗？”REVIEW CHECKS 下：四项检查各给出通过、不通过或需要补充细节的结论，例如“每个说法都可追溯到 [A1]-[A3]：通过。”

## 参考资料与复用

- [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · Reviewed 2026-10-04

TokRepo 原创提示词 · [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)。参考资料保留各自原有权利。

---
Source: https://tokrepo.com/en/workflows/speech-draft-prompt-real-anecdotes-only-b45ec831
Author: Prompt Lab